2D Image Set to 3D Space: A Novel Approach of Using Overlapping Area for Ordered and Unordered Image Set

3D reconstruction from ordered or unordered 2D image sets without additional metadata is a critical yet challenging task in fields such as architecture, archaeology, and virtual reality. Traditional methods rely on ordered image sequences with additional metadata, which are often unavailable in real-world scenarios, leading to inaccurate spatial alignments. This paper introduces a novel approach that leverages overlapping areas for robust feature detection and matching, enabling accurate 3D model reconstruction from ordered and unordered image sets. The proposed method employs a combination of SuperPoint FLANN-based matching for feature detection and feature matching. The proposed system automatically determines the spatial relationships through overlapping area detection which significantly reduces the dependency of predefined image orders. The system is implemented based on the proposed method and is tested on the COLMAP and ETH3D datasets which demonstrate that our approach improves reconstruction accuracy for ordered and unordered images without metadata.